CN104771177B - Tumble detection system and method based on machine learning - Google Patents
Tumble detection system and method based on machine learning Download PDFInfo
- Publication number
- CN104771177B CN104771177B CN201510232988.9A CN201510232988A CN104771177B CN 104771177 B CN104771177 B CN 104771177B CN 201510232988 A CN201510232988 A CN 201510232988A CN 104771177 B CN104771177 B CN 104771177B
- Authority
- CN
- China
- Prior art keywords
- fall detection
- tumble
- cloud platform
- algorithm
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Alarm Systems (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
- Emergency Alarm Devices (AREA)
Abstract
The invention discloses a tumble detection system based on machine learning. The tumble detection system based on machine learning comprises at least one tumble detector and a cloud platform; the tumble detectors are used for detecting the movement condition and/or the physiological features of a user, generating tumble judgment data according to the movement condition and/or physiological feature information and judging whether the user tumbles or not; when the user is judged to be in a tumble state, data detected by the tumble detectors are transmitted to the cloud platform to be stored. The cloud platform executes a machine learning algorithm to analyze the data detected by all the tumble detectors to judge whether update data exist, if yes, the cloud platform generates a new tumble detection algorithm sample, and meanwhile the tumble detectors generate a new tumble detection algorithm according to the new tumble detection algorithm sample. The tumble detection system based on machine learning can check whether the new tumble judgment algorithm sample exist or not in use, if yes, the new algorithm is operated, and through the close-loop mode, the tumble detecting accuracy rate of tumble detection equipment in the hand of the user can be improved all the time.
Description
Technical field
The present invention relates to fall detection field, and in particular to a kind of fall detection system based on machine learning.
Background technology
As China steps into aging society, the safety precaution in old man's daily life is increasingly becoming social concerns
Focus.According to being reported in Shanghai, only 2010 because of lethal 1983 people that falls, wherein 87.1% for over-65s the elderly,
The death rate is caused up to 77.9/10 ten thousand, it is dead because of tumble equivalent to there is 4.7 the elderlys daily.Additionally, the elderly just diagnoses a disease due to wound
In example, more than half is tumble case；It is tumble case more than 80% due to wound in inpatient.
Find that injury can be reduced to minimum degree by the rescue after Falls in Old People in 20 minutes through investigation, research,
It is the prime time of rescue.Therefore a kind of accurate fall detection technology, will give play to huge in China's aging society
Social and economic benef@；
However, fall detection is again one is rich in challenging technology, Major Difficulties are as follows：
1. it is difficult to obtain real Falls in Old People data, as fall detection method the basic data with checking is designed；
2. existing sensors for data, occurs simultaneously, it is difficult to distinguish falling to exist and between non-tumble action, easily causes erroneous judgement and leaks
Sentence；3. in worldwide, also not used for the data sample of tumble proof of algorithm, this virtually increased the difficulty of proof of algorithm
Degree, is the work of an initiative；4. the difference of everyone personal feature, such as sex, height, body weight, all can cause to fall
Some data differences of monitoring process, so as to increase the difficulty of tumble algorithm development and checking.
The content of the invention
In view of this, it is an object of the invention to provide a kind of fall detection system and method based on machine learning.
An object of the present invention is achieved through the following technical solutions, a kind of fall detection based on machine learning
System, including at least one fall detection instrument and cloud platform,
At least one fall detection instrument, for detecting the motion conditions and/or physiological characteristic of user, and according to motion
Situation and/or physiological characteristic information are generated to fall and judge data, judge whether user falls；When judging user for tumble state
When, the data that at least one fall detection instrument is detected are sent to cloud platform and are stored.
Preferably, the data that all fall detection instrument of the cloud platform execution machine learning algorithm analysis are detected judge
No presence is updated the data, and if so, then cloud platform generates new fall detection algorithm sample, while at least one fall detection instrument root
New fall detection algorithm is generated according to new fall detection algorithm sample.
Or, the cloud platform performs machine learning algorithm and analyzing the data that all fall detection instrument detect and judging whether to deposit
Updating the data, if so, then cloud platform generates new fall detection algorithm sample, while cloud platform is calculated according to new fall detection
Method sample generates new fall detection algorithm, and the new fall detection algorithm for generating is sent to fall detection instrument by cloud platform, is made
For the algorithm of judgement of falling next time.
The second object of the present invention is achieved through the following technical solutions, a kind of fall detection side based on machine learning
Method, including：The motion conditions and/or physiological characteristic of step a. fall detection instrument detection user, and according to motion conditions and/or life
Reason characteristic information is generated to fall and judges data, judges whether user falls.
Preferably, when user is judged for tumble state, the data that fall detection instrument is detected are sent to cloud platform and carry out
Storage.
Preferably, when cloud platform receives the data that fall detection instrument is detected, cloud platform performs machine learning algorithm
Analyze the data to judge whether to update the data；If so, then cloud platform generates new fall detection algorithm sample, while falling
Detector generates new fall detection algorithm according to new fall detection algorithm sample；Or the cloud platform performs machine learning and calculates
The data that method analysis fall detection instrument is detected judge whether to update the data, and if so, then cloud platform generates new tumble inspection
Method of determining and calculating sample, while cloud platform generates new fall detection algorithm, fall detection instrument according to new fall detection algorithm sample
New fall detection algorithm is obtained from cloud platform, as the algorithm of judgement of falling next time.
As a result of above-mentioned technical proposal, the present invention has the advantage that：
1. the present invention is by the way that tumble alert data is stored, and increasing with user, can set up one increasingly
Complete tumble sample database；
2. Yunan County's platform updating the data in database, performs machine learning algorithm, and adjust automatically operates in tumble
Tumble evaluation algorithm on alarm, creates new tumble evaluation algorithm sample.
3 have the personal information that records each user in platform, such as：Sex, age, height, body weight etc., with sample number
It is continuously increased according to the record in storehouse, machine learning algorithm combines a certain degree of manual analysis, can be right according to these personal information
The Different Effects of tumble process sensor data, such that it is able to reach the target for personalized fall detection algorithm.Allow and use
Person, the fall detection algorithm for having customization, so as to reach the target for improving accuracy of detection.
4. electricity on each fall detection instrument, can check whether there is new tumble evaluation algorithm sample, if any meeting automatically from clothes
Business device down loading updating, by this closed loop mode, user's fall detection equipment on hand detects that the accuracy rate of tumble can be carried always
It is high.
Description of the drawings
In order that the object, technical solutions and advantages of the present invention are clearer, below in conjunction with accompanying drawing the present invention is made into
The detailed description of one step, wherein：
Fig. 1 is the theory diagram based on the fall detection system of machine learning；
Fig. 2 is the theory diagram of fall detection instrument；
Fig. 3 is blood pressure detecting unit construction principle block diagram；
Fig. 4 is temperature monitoring unit construction principle block diagram；
Fig. 5 is fall detection instrument workflow diagram；
Fig. 6 is cloud platform workflow diagram；
Fig. 7 is to fail to report data reference；
Fig. 8 is the reference of correct alert data；
Fig. 9 is wrong report data reference.
Specific embodiment
Below with reference to accompanying drawing, the preferred embodiments of the present invention are described in detail；It should be appreciated that preferred embodiment
Only for the explanation present invention, rather than in order to limit the scope of the invention.
As shown in figure 1, a kind of fall detection system based on machine learning, including at least one fall detection instrument and Yun Ping
Platform,
At least one fall detection instrument, for detecting the motion conditions and/or physiological characteristic of user, and according to motion
Situation and/or physiological characteristic information are generated to fall and judge data, judge whether user falls；When judging user for tumble state
When, the data that at least one fall detection instrument is detected are sent to cloud platform and are stored.
The cloud platform performs machine learning algorithm to be analyzed the data that all fall detection instrument detect and judges whether
Update the data, if so, then cloud platform generates new fall detection algorithm sample, while at least one fall detection instrument is according to new
Fall detection algorithm sample generates new fall detection algorithm.
Or, the cloud platform performs machine learning algorithm and analyzing the data that all fall detection instrument detect and judging whether to deposit
Updating the data, if so, then cloud platform generates new fall detection algorithm sample, while cloud platform is calculated according to new fall detection
Method sample generates new fall detection algorithm, and the new fall detection algorithm for generating is sent to fall detection instrument (this by cloud platform
In fall detection instrument refer to detect tumble shape or be not detected by the fall detection instrument of tumble state), as next
The secondary algorithm for judging of falling；Certainly in the present embodiment, fall detection instrument can also automatically obtain at set intervals renewal
Fall detection algorithm sample.
The fall detection system based on machine learning also includes mobile terminal, when fall detection instrument judges that user falls
Alert messages are generated during state and the alert messages is sent to mobile terminal.
As shown in Fig. 2 the fall detection instrument is obtained including what micro-control unit and micro-control unit connected for real-time
The 3-axis acceleration sensor of user's body attitude；The height for acquisition alarm height in real time being connected with micro-control unit
Sensor；The GPS module for user in real positional information being connected with micro-control unit；It is connected with micro-control unit
For obtaining the BLE modules of biology sensor value；The GPRS module for transmission data being connected with micro-control unit；And be
The battery that power unit is powered；The startup of the micro-control unit controls fall detection instrument and receive user body posture data,
And judgement is analyzed to attitude data, and alarm unit is started according to judged result.
The biology sensor includes blood pressure monitoring unit and temperature monitoring unit.
As shown in figure 3, the blood pressure monitoring unit includes light emission module, Optical Receivers and pulse wave processing module；
The light emission module, under the control of micro-control unit, to user dual wavelength light is launched；The Optical Receivers, receives
The dual wavelength light of user's reflection, and the optical signal for receiving is converted to into electric signal, and by the electric signal transmission to pulse
Ripple processing module；The pulse wave processing module, is amplified to electric signal, filters, secondary amplification and analog-to-digital conversion, concurrently
The warp let-off amplifies the information after filtering analog-to-digital conversion to micro-control unit.
As shown in figure 4, the temperature monitoring unit includes temperature sensor and processes temperature signal module；The temperature is passed
Sensor, gathers and sends user's body temperature information to processes temperature signal module；The processes temperature signal module, receives temperature and passes
User's body temperature information of sensor collection, is amplified to user's temperature information and filters and analog-to-digital conversion, and the concurrent warp let-off is amplified
Information after filtering analog-to-digital conversion is to micro-control unit.
A kind of fall detection method based on machine learning, including：
The motion conditions and/or physiological characteristic of step a. fall detection instrument detection user, and according to motion conditions and/or life
Reason characteristic information is generated to fall and judges data, judges whether user falls；When user is judged for tumble state, fall detection instrument
Information warning is generated, while the data that fall detection instrument is detected are sent to cloud platform being stored.When fall detection instrument judges
Alert messages are generated when user is tumble state and the alert messages is sent to mobile terminal.
In the present embodiment, letter people's information of each user is previously stored in cloud platform, such as：Sex, the age, height,
The related physiological data such as body weight and the fall detection algorithm sample obtained according to the data calculating of storage, including threshold value.When
When cloud platform receives the data that fall detection instrument is detected, cloud platform confirms whether data are correctly reported to the police, report by mistake, failed to report, such as
Shown in Fig. 7～9.
Wrong report refers to that the data that fall detection instrument is detected compare with the data in cloud platform, and difference exceeds set threshold
Value.
Fail to report and refer to, fall detection instrument judges user for tumble state, but cloud platform is not received by the number of this time detection
According to.
When cloud platform receives the data that fall detection instrument is detected, cloud platform performs machine learning algorithm and analyzes the number
It is judged that whether there is updating the data；If so, then cloud platform generates new fall detection algorithm sample, while fall detection instrument root
New fall detection algorithm is generated according to new fall detection algorithm sample, the detection algorithm is used as judging whether user falls next time
Algorithm.
In the present embodiment, the fall detection algorithm is：The data that fall detection instrument is detected are cut through sliding window
Take, be calculated feature value vector x, Jing after normalization as network inputs, such as (2.1) are shown for this feature value vector,
X=[x0,x1,x2,…,x7] (2.1)
Wherein x0=-1；x1=amaxFor acceleration and maximum；x2=aminFor acceleration and minimum of a value；x3
=Δ t is the time difference of maxima and minima；x4=aσFor acceleration and variance；For acceleration peace
Average；For n before an X-direction mean value；For the mean value of n after X-direction,
Then i-th node of hidden layer is input into weights as shown in (2.2),
Then i-th node is output as：
yi=f (xwi) (2.3)
Wherein f (m) is activation primitive, and activation primitive f (m) adopts sigmoid function,
Then hidden layer exports y=[y1,y2,y3], output layer weight w=[w1,w2,w3], final output is：
P=f (yw) (2.4)
Judged by p value, if fall, when p value is more than threshold value, be judged as falling, when p value is less than or equal to threshold value,
It is judged as non-tumble.
In the present embodiment, illustrate by taking 10 fall detection instrument as an example.Such as, wherein 5 fall detection instrument detections
To tumble state, now, the data that 5 fall detection instrument will be deemed as corresponding to tumble state are sent to cloud platform and are deposited
Storage, the data that all fall detection instrument of cloud platform execution machine learning algorithm analysis are detected judge whether to update the data,
If so, then cloud platform generates new fall detection algorithm sample, while cloud platform is generated according to new fall detection algorithm sample
The new fall detection algorithm for generating is sent to (the tumble inspection here of fall detection instrument by new fall detection algorithm, cloud platform
Survey instrument and refer to detect tumble shape or be not detected by the fall detection instrument of tumble state), judge as falling next time
Algorithm.
In the present embodiment, new fall detection algorithm can on the basis of former fall detection algorithm, change threshold value
Size, be allowed to in fall detection algorithm sample in cloud platform threshold value be adapted, it should be noted that in cloud platform generate
Fall detection algorithm sample in threshold value be change.Certainly the threshold value in fall detection instrument in fall detection algorithm can also
Identical with the threshold value in fall detection algorithm sample, the threshold value in fall detection algorithm sample is by obtaining on fall detection instrument
The data of biography are calculated and obtained.
In the present embodiment, new fall detection algorithm can also be the characteristic vector by changing former fall detection algorithm
Value, such as original characteristic vector value x=[x0,x1,x2,…,x7], and new characteristic vector value is x'=[x0,x1,
x2,......,x5,x6,x7,x8]。
Table 1 is the contrast of fall detection algorithm final detection result twice after machine learning algorithm：
As seen from the above table, after machine learning, the algorithm of two versions, rate of failing to report is reduced：52.1%；Rate of false alarm drops
It is low by 6.8%；
The preferred embodiments of the present invention are the foregoing is only, the present invention is not limited to, it is clear that those skilled in the art
Member the present invention can be carried out it is various change and modification without departing from the spirit and scope of the present invention.So, if the present invention
These modifications and modification belong within the scope of the claims in the present invention and its equivalent technologies, then the present invention is also intended to comprising these
Including change and modification.
Claims (3)
1. a kind of fall detection method based on machine learning, it is characterised in that：Including
The motion conditions and/or physiological characteristic of step a. fall detection instrument detection user, and it is special according to motion conditions and/or physiology
Reference breath is generated to fall and judges data, judges whether user falls；
When user is judged for tumble state, the data that fall detection instrument is detected are sent to cloud platform and are stored step b.；
The fall detection algorithm is：
The data detected by fall detection instrument, are calculated feature value vector x, as net after the vector normalization of this feature value
The input layer of network, such as shown in (2.1),
X=[x0,x1,x2,…,x7] (2.1)
Wherein x0=-1；x1=amaxFor acceleration and maximum；x2=aminFor acceleration and minimum of a value；x3=Δ t
For the time difference of maxima and minima；x4=aσFor acceleration and variance；For acceleration and mean value；For n before an X-direction mean value；For the mean value of n after X-direction,
Then i-th node of hidden layer is input into weights as shown in (2.2),
Then i-th node is output as：
yi=f (xwi) (2.3)
Wherein f (m) is activation primitive, and activation primitive f (m) adopts sigmoid function,
Then hidden layer exports y=[y1,y2,y3], output layer weight w=[w1,w2,w3], final output is：
P=f (yw) (2.4)
Judged by p value, if fall, when p value is more than threshold value, be judged as falling, when p value is less than or equal to threshold value, judged
For non-tumble.
2. the fall detection method based on machine learning according to claim 1, it is characterised in that：When cloud platform is received
During the data that fall detection instrument is detected, cloud platform execution machine learning algorithm analyzes the data and judges whether to update number
According to；If so, then cloud platform generates new fall detection algorithm sample, while fall detection instrument is according to new fall detection algorithm sample
The new fall detection algorithm of this generation.
3. the fall detection method based on machine learning according to claim 1, it is characterised in that：The cloud platform is performed
The data that machine learning algorithm analysis fall detection instrument is detected judge whether to update the data, and if so, then cloud platform is generated
New fall detection algorithm sample, while cloud platform generates new fall detection algorithm according to new fall detection algorithm sample,
The new fall detection algorithm for generating actively is sent to fall detection instrument by cloud platform, used as the algorithm of judgement of falling next time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510232988.9A CN104771177B (en) | 2015-05-08 | 2015-05-08 | Tumble detection system and method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510232988.9A CN104771177B (en) | 2015-05-08 | 2015-05-08 | Tumble detection system and method based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104771177A CN104771177A (en) | 2015-07-15 |
| CN104771177B true CN104771177B (en) | 2017-05-10 |
Family
ID=53613183
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510232988.9A Active CN104771177B (en) | 2015-05-08 | 2015-05-08 | Tumble detection system and method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104771177B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105125220A (en) * | 2015-10-20 | 2015-12-09 | 重庆软汇科技股份有限公司 | Falling-down detection method |
| CN105769205A (en) * | 2016-02-23 | 2016-07-20 | 中国科学院深圳先进技术研究院 | Body information detection device and fall detection system |
| EP3522782A4 (en) * | 2016-10-05 | 2020-05-13 | MY Medic Watch Pty Ltd | Alert system |
| CN110139598B (en) * | 2016-12-29 | 2022-07-01 | 关健强 | Monitoring and tracking system, method, article and apparatus |
| CN108771543B (en) * | 2018-04-16 | 2020-11-03 | 齐鲁工业大学 | Old man falling detection method and system in real environment based on big data |
| CN108606778B (en) | 2018-04-16 | 2024-04-19 | 顾继炎 | Medical device, algorithm updating method, medical system and external monitoring device |
| CN108634968A (en) * | 2018-04-26 | 2018-10-12 | 深圳市科思创动科技有限公司 | health monitoring system and method |
| TW202042739A (en) * | 2019-05-29 | 2020-12-01 | 友達光電股份有限公司 | Physiological information warning system and physiological information warning method |
| CN112885035A (en) * | 2021-01-13 | 2021-06-01 | 厦门米蓝信息科技有限公司 | Old man falling detection method and system in real environment based on big data |
| CN115530774B (en) * | 2021-06-30 | 2024-03-26 | 荣耀终端有限公司 | Epilepsy detection method and device |
| CN117414119A (en) * | 2023-10-18 | 2024-01-19 | 北京大学国际医院 | Noninvasive intracranial pressure estimation device for assisting in measuring diameter of optic nerve sheath by artificial intelligence |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2005237576A (en) * | 2004-02-25 | 2005-09-08 | Kansai Electric Power Co Inc:The | Tumble judgment device |
| CN102132331B (en) * | 2008-08-28 | 2014-09-24 | 皇家飞利浦电子股份有限公司 | fall detection and/or prevention system |
| CN102136180B (en) * | 2011-03-11 | 2013-09-18 | 北京航空航天大学 | Device for detecting and alarming human body tumble |
| CN103294879A (en) * | 2012-02-24 | 2013-09-11 | 中国科学院沈阳计算技术研究所有限公司 | Portable remote health monitoring system and implementation method thereof |
| CN104361361B (en) * | 2014-11-14 | 2018-04-03 | 北京天地弘毅科技有限公司 | The method and system for judging to fall down by cloud computing and machine learning algorithm |
-
2015
- 2015-05-08 CN CN201510232988.9A patent/CN104771177B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN104771177A (en) | 2015-07-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104771177B (en) | Tumble detection system and method based on machine learning | |
| CN105590409B (en) | A kind of tumble detection method for human body and system based on big data | |
| CN103886715B (en) | A kind of tumble detection method for human body | |
| Pingale et al. | Disease prediction using machine learning | |
| CN109397703B (en) | Fault detection method and device | |
| CN105139029B (en) | A kind of Activity recognition method and device of prison prisoner | |
| Xefteris et al. | Performance, challenges, and limitations in multimodal fall detection systems: A review | |
| Hossain et al. | A direction-sensitive fall detection system using single 3D accelerometer and learning classifier | |
| WO2009071598A1 (en) | Method and apparatus for detecting a critical situation of a subject | |
| US11042768B1 (en) | Deep learning-based object recognition system and method using PIR sensor | |
| CN111919242B (en) | System and method for processing multiple signals | |
| CN106650300B (en) | Old man monitoring system and method based on extreme learning machine | |
| Shi et al. | Fall detection system based on inertial mems sensors: Analysis design and realization | |
| Ramachandran et al. | Evaluation of feature engineering on wearable sensor-based fall detection | |
| WO2021183301A1 (en) | Systems and methods for non-invasive virus symptom detection | |
| CN107019501B (en) | Remote tumble detection method and system based on genetic algorithm and probabilistic neural network | |
| CN111317458A (en) | Blood pressure detection system based on deep learning | |
| Qu et al. | Convolutional neural network for human behavior recognition based on smart bracelet | |
| CN111232778B (en) | Method and device for counting number of people in elevator car | |
| Poorani et al. | Sensor based decision making inference system for remote health monitoring | |
| CN112767202A (en) | Remote monitoring platform and monitoring method | |
| Cheng et al. | A fall detection algorithm based on pattern recognition and human posture analysis | |
| CN116763299A (en) | Fall detection system and method based on wearable sensor | |
| Khawandi et al. | Applying machine learning algorithm in fall detection monitoring system | |
| KR20230092400A (en) | vulnerable social group danger recognition detection method based on multiple sensing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20230626 Address after: 519085 Zone D, 4th Floor, No. 1 Jinyuan 1st Road (Plant 1), Jinding Industrial Zone, High tech Zone, Zhuhai City, Guangdong Province Patentee after: ZHUHAI KADEN MEDICAL IMAGING TECHNOLOGY Co.,Ltd. Address before: No. 73-14-7, Keyuan 1st Road, Jiulongpo District, Chongqing 400039 Patentee before: CHONGQING RUANHUI TECHNOLOGY CO.,LTD. |
|
| TR01 | Transfer of patent right |